E
u r o p e a n
E
x p e r t
N
e t w o r k o n
E
conomics of
Education
(EENEE)
EENEE Analytical Report No. 10
Prepared for the European Commission
Torberg Falch and Hessel Oosterbeek
September 2011
European Commission Education and Culture
10
Financing lifelong learning: Funding
mechanisms in education and training
Financing lifelong learning:
Funding mechanisms in education and training*
Torberg Falch
Norwegian University of Science and Technology and CESifo Hessel Oosterbeek
University of Amsterdam, Tinbergen Institute, and CESifo
Abstract
This report provides an overview of available evidence on funding mechanisms in education and training. The focus is on empirical evidence relevant for European education institutions. The report covers important research issues for primary and secondary education, higher education, and adult learning. We draw conclusions on the advantages and drawbacks of a wide range of funding mechanisms, including tax incentives, grants and subsidies, loans, and individual learning accounts. The advantages of funding schemes are evaluated with respect to efficiency, quality, equity, and sustainability.
JEL-codes: H23; H7; I2; J24;
Keywords: Lifelong learning; Education finance; Education performance
*
Section 2 and 3 of this report are prepared by Torberg Falch while Section 4 is prepared by Hessel Oosterbeek. We thank representatives for DG EAC and CEDEFOP and Lars-Erik Borge for valuable comments. The views and conclusions in this report
Content
Executive summary ... 1
1. Introduction ... 4
2. Financing primary and secondary education ... 4
2.1. Demand functions for education spending ... 5
2.1.1. Grants ... 6
2.1.2. Demographics... 8
2.2. Allocation of education resources ... 10
2.3. Centralized vs. decentralized finance ... 12
2.3.1. Education spending ... 12
2.3.2. Education performance... 14
2.4. Financial incentives for students ... 16
2.5. Concluding remarks... 17
3. Financing higher education ... 18
3.1. Tuition fees ... 19
3.2. Student aid ... 20
3.3. Student loans... 21
3.4. Concluding remarks... 22
4. Funding mechanisms for adult learning ... 24
4.1. Evidence-based policies ... 24
4.2. Results ... 25
4.2.1. Subsidies... 25
4.2.2. Vouchers... 27
4.2.3. Individual Learning Accounts ... 27
4.2.4. Tax instruments ... 29
4.4. Concluding remarks... 32
5. Conclusions ... 33
References ... 34
Opinions expressed in this report are those of the authors alone and do not represent the point of view of the European Commission.
Executive summary
This report presents major findings in the empirical literature on funding mechanisms in education and training. The focus is on evidence relevant for efficient and equitable provision within European education institutions. The report covers primary, secondary and higher education in addition to adult education and training.
Primary and secondary education
Centralized funding contributes to more equal education spending across regions and local governments than decentralized funding. On the other hand, decentralized funding seems to provide better incentives to improve school quality. In particular property taxation seems to be a discipline device for school providers, probably because school quality capitalizes into house prices and thereby increases property tax income. There is potentially an efficiency-equity trade-off in the choice of finance scheme.
The main funding mechanism with centralized finance and local provision is grants. The evidence indicates that grants are mainly used on public spending and not on reduced local tax rates; “the money stick where it hits”. This “flypaper effect” is essential for centralized finance to equalise spending; grants are not mainly used on tax reliefs. The size of the flypaper effect, however, depends on the degree of tax discretion for local governments. Thus, the efficiency-equity trade-off in the choice of finance scheme is not easily avoided since the leakage of grants to tax reliefs seems to be highest when local governments have high tax setting discretion.
Another potential discipline device is financial incentives to students. It does not exist much evidence for primary and secondary education in industrialized countries, but all the evidence clearly indicates that such incentives improve student performance.
Optimal school size in terms of spending is large and about the size of the largest schools observed. However, the importance of school size for student performance is mixed. Some evidence indicates that small schools have lowest performance even in cases where school consolidations imply less competition across schools.
Demographics affect the demand for education spending. One may be concerned about the willingness and possibility to fund education in an aging society. The empirical evidence on the demand for education spending suggests that such a concern is misplaced. Even though the evidence clearly indicates that a larger share of elderly in society decreases education spending per student, the evidence also shows that a smaller share of students in the population increases spending per student. Since a larger share of elderly and a smaller share of students in most cases go hand-in-hand in an aging society, the effect on spending per student will tend to cancel out.
Higher education
Higher education is investment in future wage premiums. It is heavily subsidized in Europe despite the fact that the private return to higher education is significant.
We review the literature on how students react on changes in tuition fees and student aid. Evidence on student behaviour is critical in order to judge different arguments for subsidizing higher education to the present extent. The typical finding is that rising annual tuition fees with 1000 euro or reducing student aid with 1000 euro decreases enrolment in higher education by about 5 percentage points. The responses to aid and tuition are of about the same magnitude, and the evidence indicates that student behaviour in this respect is independent of socioeconomic background. The results are consistent with students considering education as an investment and behave rationally. Public student loan systems will provide liquidity to the students and overcome inefficiencies in private provision. The systems with the best equity features are income contingent loans and graduate taxes. With income contingent loans, repayment depends on achieved income. With graduate taxes, students receive a grant that they repay as a special tax on income after graduation.
The literature includes several arguments for subsidizing higher education. The least controversial argument is the income tax-wedge. Regarding all other arguments it is possible to question whether they are real at all. The best argument is probably that higher education might foster dissemination of knowledge. It is however difficult to quantify the size of such externalities, and it may even be argued that they are very limited. Overall, the literature clearly shows that the costs should be shared between the governments and the individual students. In addition, decentralizing decisions on tuition fees to the higher education institutions may enhance competition as a discipline device.
Adult learning
In line with the approach in Schlotter et al. (2009), we focus on evidence that comes from studies that make use of a credible comparison group. Doing so reveals that the available evidence is very fragmented. That is, however, not due to our approach but is the current state of knowledge. Not hiding this, for instance by over-interpreting results from studies that do not meet the standards, makes clear that more and often a different type of research is needed.
To stimulate lifelong learning activities of adult workers, it is often believed that simply making such activities available at a low price provides insufficient incentives. Instruments such as vouchers and individual learning accounts give potential learners a very explicit confirmation of their increased purchasing power. This should strengthen people’s awareness of the availability and importance of learning activities. Evidence
from England confirms that simply making training available at a low price (or even for free) indeed does not provide sufficient incentives, at least not to the low skilled employees at which this program was targeted. Perhaps even more surprising is that this is the case despite the fact the program included an element of information, advice, and guidance to employers and employees.
The most convincing evidence regarding vouchers comes from a policy targeted towards American veterans. The evaluation finds significantly positive effects of vouchers on college attainment. It is questionable, however, to what extent the results would carry over to the current situation in Europe. After all, the policy took place more than 35 years ago, veterans are a rather specific group, and the education system in the US is markedly different from that in Europe.
In Switzerland and the Netherlands, interesting experiments have been conducted to measure the effects of individual learning accounts. Although the experiments differ in the people being eligible, the studies find a remarkably similar inefficiency. The studies conclude that more than half of the redeemed vouchers used for training would have been paid by the individuals (or their employers) themselves.
A less explicit way of subsidizing learning activities is in the form of tax deductions. Existing evidence is restricted to the Netherlands. An age dependent tax deduction for firms appears to have led to postponement of training activities rather than to an increase. An income tax deduction for individuals appears to have substantially positive effects on training participation. This may be due to the fact that this instrument is available to everyone while the other policies are typically aimed at specific groups of low skilled workers.
1. Introduction
Education and training is an important policy area. In economic terms, about 11 percent of total public expenditures are used on education on average across the EU countries (EU, 2009), ranging from below 10 percent in Germany, Italy and Luxembourg to above 15 percent in Cyprus and Denmark. This review discusses education finance issues for education and training; from compulsory schools to adult learning.1The focus is on empirical evidence relevant for European education institutions. Our interest is on efficient and equitable provision of education and training. The concept of efficiency relates inputs and outputs in a production process. In our context the inputs of interest are funding mechanisms of education and training, while relevant output measures include student achievement, educational attainment, employment opportunities, and wages.
Research questions in the literature are extensive and typically specific to each educational level. For primary and secondary education, our focus is on financing systems and demand for spending. For higher education, we cover the issue of private versus public funding. Adult training is at the outset a private responsibility, and our focus will be on the effects of subsidies and tax incentives. All these issues are lively research areas, and we concentrate on studies that make use of carefully constructed comparison groups (cf. Schlotter et al., 2009).
2. Financing primary and secondary education
Primary and secondary education can be financed in many different ways. When education is the responsibility of the central government, it is financed directly by national tax income. In most countries, however, local governments are at least partly responsible for this educational level. Then the financing can be based on local taxes or on grants from the central government. The grants can take two principally different forms; they are either unconditional or conditional on local government decisions. Conditional grants are typically related to how much the local government spend. The typical setting is that local governments are financed by all these means.
Decisions on education spending are taken by elected politicians in political institutions. Citizens’ preferences for educational services and education spending are transformed into actual policies through some kind of political processes. In order to understand aspects of these processes empirically, a theoretical framework to guide specifications
1
Early childhood education prior to compulsory education is not captured by this review. For surveys of the literature on economic impacts of early childhood education, see for example Wössmann (2008) and Heckman et al. (2010).
of empirical models would be helpful.2 In the lack of stringent theoretical models, however, empirical papers typically confine themselves to establish empirical regularities. Such empirical models include important elements in the budget constraint of governments in addition to various demographic characteristics.
The main interest is in causal effects on education spending and on student achievement. Revealing causal effects requires careful empirical designs as explored in, e.g., Schlotter et al. (2009). Policy interventions, which have become standard to exploit in order to estimate causal effects in micro-econometric models, are however not used extensively in public finance. One is often interested in understanding the institutions deciding on interventions, and then the interventions cannot be regarded as exogenous. Several papers use a difference-in-difference approach instead based on panel data. For example, panel data on local governments include information for several years, and the difference-in-difference approach will difference out possible important unobserved and hard-to measure factors, such as political tastes and geography. Another way to circumvent the potential flaws of unobserved variables has been to study responses of local governments to interventions decided by the central government. One example of such interventions is grant reforms. Finally, some papers use the instrumental variable approach in order to identify the casual effects of interest on a particular subset of the variation in data, i.e., variation that can be considered as exogenous.
This section firstly discusses results from demand functions for education spending, including responses to changes in income, grants, and demographics. Thereafter we present evidence on the allocation of resources across schools. Thirdly, since local governments are responsibility for primary and secondary education in most countries, centralized financing of education spending via grants vs. decentralized financing by local taxes is discussed.
2.1. Demand functions for education spending
The first studies on demand functions in education utilized cross-section variation across school districts to estimate income elasticities, grant effects, and some other parameters. The typical income elasticity estimated is about 0.5, see for example
2
The workhorse model to understand public sector decision making has been the median voter model in which the outcome is in accordance with the preferences of one decisive voter. Bergstrom and Goodman (1973) specify a set of assumptions in order to estimate demand functions for public sector services in the median voter framework. One of the critical assumptions is that the decision problem is unidimensional. Reviews of the appropriateness of the median voter model as a starting point of empirical work include Craig and Inman (1986) and Rubinfeld (1987). While the median voter model may be reasonable for school districts in the US which only purpose is to provide educational services, it is more problematic in a European setting since education is typically provided by multi-purpose governments. Political decisions involve bargaining within legislatures and with interest groups, restricted by institutional characteristics (Romer et al., 1992, Von Hagen, 1992). In addition, spending decisions may involve a common pool problem in which the benefits are concentrated to specific geographical areas (local
Ehrenberg (1973) and Feldstein (1975). The result implies that if the income increases by, say, 10 percent, education spending increases by 5 percent. Wealthy districts have higher education spending than poor districts, but the difference in education spending is smaller than the difference in private income. Later studies using cross-section data confirms that education is a normal good with income elasticity below unity.
Studies estimating dynamic models of education spending find higher income elasticities. Falch and Rattsø (1997) investigate the development of spending per student in primary education in Norway in the period 1880-1990. They use aggregate data and distinguish between spending components decided at the central and at the local level. They find that teacher wages is positively related to GDP per capita and that in particular the number of teachers per class responds strongly to the long run growth in GDP. The income elasticity with respect to total spending adds up to above two. Heinesen (2004) estimate a dynamic panel data model for Danish local governments for the period 1984-1996. He finds a long run income elasticity of about 1.3 when income is measured as the sum of private income and grants. However, the effect of grants is much larger than the effect of private income.
2.1.1. Grants
In most countries, grants are the main income source of local governments. It is an important distinction between lump-sum grants and matching grants. Lump-sum grants, often called block grants, are unconditional for the local government. Decisions on how to spend the grants are independent decisions. Lump-sum grants can be used on education or other local public services. In addition, increased lump-sum grants can be used to reduce local taxes. In fact, economic theory based on traditional utility maximization behaviour implies that one additional euro in private income and one additional euro in lump-sum grants have exactly the same impact on spending on local public services. In the former case the number of euro taxed will increase while in the latter case it will decrease, but in such ways that private after-tax income will be the same. A critical assumption for this result is that there is some degree of local tax discretion.
Matching grants are conditional on some specific local government behaviour. They affect the relative prices facing the local government. Matching grants typically pays the local government a share of the spending on a particular service.
The usual finding in the local public finance literature is that the effect on spending is larger of increased lump-sum grants than of increased private income. This is labelled the flypaper effect; “money stick where it hits”. Early contributions are Pommerehne and Schneider (1978) on Swiss data and Wyckoff (1991) on Michigan school districts data. Hines and Thaler (1995) and Inman (2008) present overviews of the flypaper literature.
More recent contributions have forcefully argued that grants are often endogenous in the sense that local governments can influence the grant they get. Traditional models do not take into account that grants can be decided in negotiations between central and local politicians and that some unobserved local socio-economic characteristics might affect both how the grants are allocated across local governments and the spending and taxation decisions.
Ladd (1993) appears to be the first attempt to overcome the potential endogeneity problem. She exploits a change in the federal taxation rules in the US in 1986 as a natural experiment. One part of the tax reform was to broaden the federal tax base. This implied a significant revenue windfall for the states which used the federal tax base for their own income taxes. Without any flypaper effect, the states would reduce the tax rate such that private consumption and state spending did not change. However, Ladd finds a significant flypaper effect. One additional dollar in tax windfalls led states to retain 40 cents more than they have in the absence of the windfalls.
Evans and Zhang (2007) utilize that many US states with state lotteries earmark lottery profits for primary and secondary education. Using the yearly variation in lottery profits, they find that 50-70 percent of an additional dollar in earmarked lottery profits is transferred to local school districts, and that spending per student increases by 80-100 percent of these grant windfalls. Gordon (2004) investigates the effect of exogenous changes in the US federal grant to primary and secondary education. The grant is determined by decennial census data on child poverty which changes every 10 years. The grant level for the individual school districts changed in 1993 as a result of new census data available. Exploiting this discontinuity, she finds that increased grants boost total instructional spending about dollar for dollar the first year. Three years later, however, the local revenues seem to have declined considerably, and the instructional spending is not significantly different from the initial level.
Another study that also finds limited support for the flypaper effect is Lutz (2010). He exploits a school finance reform in 1999 in New Hampshire that increased state grants by about 200 percent, a reform mandated by a New Hampshire Supreme Court decision. New Hampshire has no state-imposed limitations on the taxing or spending behaviour of local governments, and school districts budgets are set directly by voters in referendums and not by elected representatives. Lutz (2010) finds that about 90 cents per grant dollar are spent on tax reduction and only 10 cents on education.
Few European studies on causal effects of grants seem to exist. The exception is some studies from Sweden. Dahlberg et al. (2008) exploit a discontinuity in the Swedish grant system. Local governments with out-migration rate above a certain level receive extra
grants. Identifying grant effects on this discontinuity, they find that federal grants increase local spending about one-to-one without any effect on local taxes.
Overall, it seems like there is a flypaper effect in most cases, but that there are instances where higher grants are used to mainly reduce taxes such that increased grants have about the same effect on education spending as increased private income. The evidence indicates that the size of the flypaper effect is related to local political institutions. Dahlberg et al. (2008) argue that flypaper effects can be “rational” when the central government has better tax instruments than local governments. Other studies have found that the size and existence of flypaper effects depend on the political and bureaucratic structure of the local governments (Strumpf, 1998, Tovmo and Falch, 2002). In addition, one would expect to find flypaper effects when there are important restrictions on local tax policy.
2.1.2. Demographics
Another central issue in demand functions for education spending has been the effects of demographic factors. An aging population will increase the political pressure to tilt the composition of social spending in favour of the elderly, which potentially sacrifice on education. On the other hand, the number of students per capita is a cost factor in educational production. When the student share increases, it becomes more expensive to provide a given school quality. The young and the old “compete” for scarce public sector resources.
Poterba (1997) examines total spending on primary and secondary education at the US state level in a panel data framework. States that experience high growth in the population above 65 years appear to have lower growth in education spending than other states. All else equal, states with more elderly spend less on public schools. Regarding the share of the population of school-age children, Poterba finds a very strong negative effect on spending per student. His results indicate that spending per capita is almost independent of the number of students. Harris et al. (2001) estimate a similar model using school district data instead of state level data. They find smaller effects than Poterba (1997) both for the share of elderly and the share of school-age population.
There are a number of reasons why elderly may support public education. The elderly are altruistic to some degree, better education are likely to be capitalized into the value of the homes, and quality education may boost economic growth which in turn helps finance public pension schemes and health care. Brunner and Baldson (2004) use survey data from California in an attempt to distinguish between potential mechanisms. Similar to the findings of Poterba (1997) and Harris et al. (2001), they find that older voters relative to younger voters are more willing to support local spending on schools than state spending. They also find that homeowners to a larger degree than others support
local spending but not state spending, in accordance with the hypothesis that capitalization of school quality into housing values motivates elderly to vote for education spending. The latter difference in support is most common for elderly, which is in accordance with the altruistic behaviour hypothesis.
A negative effect of the share of students in the population on spending per student is the typical finding also in European studies. That is the case in panel data analyses of Norwegian primary and lower secondary education (Borge and Rattsø, 1995) and upper secondary education (Falch and Rattsø, 1999), general education in Russia (Verbina and Chowdhury, 2004), and for primary and lower secondary education in Denmark (Heinesen, 2004, Borge and Rattsø, 2007) and Switzerland (Grob and Wolter, 2007). These results clearly suggest that it is a disadvantage to be in a large cohort. For example, using data for Swiss cantons during the period 1990-2002, Grob and Wolter (2007) find that when the share of school-aged population decreases by 10 percent (about 1 percentage point), spending per student increases by 4 percent.
The European studies find in addition typically that the share of elderly decreases spending per student. For example Grob and Wolter (2007) find about the same numerical effect of the share of retired people as the share of school-aged population. Kempkes (2010) analyses a case that has close similarities with a natural experiment. After the German unification, the fertility rate dropped considerably in East Germany, but did not change much in western Germany. The number of primary school students in East Germany dropped by 50 percent from 1993 to 2002. This is clearly a large shock to the education system. Kempkes find a relatively fast adjustment in spending per student. The elasticity of spending per student with respect to the number of students during this period is estimated to about -0.27, which implies that a fall in the number of students by 50% decreases total spending by 36.5 percent. This is a stronger response in total spending than the typical estimate in the literature predicts. In particular, the number of full-time equivalent teachers responded strongly to the fall in the number of students, while spending on school infrastructure adjusted more sluggish.
Overall, the evidence shows that education spending does not adjust proportionately to changes in the student cohort size. Small cohorts experience higher spending per student than large cohorts, all else equal. Most studies find that education spending adjusts slowly to the change in the number of pupils, but the evidence from the East German shock in the number of students indicates that the point estimates must be interpreted with care. The evidence also shows that aging of the population works in the opposite direction. More elderly will put pressure on spending per student.
2.2. Allocation of education resources
Set aside the political processes determining overall spending on education, how are resources distributed across schools, and how should they be distributed? Interesting questions include: What does it cost to deliver education of a certain quality? What does it cost to bring specific demographic groups to a certain level of achievement? In economic terminology, such questions can be addressed by knowledge about the “cost function”. A cost function is defined as the minimum cost necessary for production of every output level. The concept is originally developed for profit-maximizing firms. For example Duncombe and Yinger (2005, 2011) claim to estimate an educational cost function for education. Unfortunately, it is hard to imagine the possibility of estimating reliable cost functions for education (Costrell et al., 2008, Falch et al., 2008). Education production is as yet only partly understood, and important aspects of the production technology in education remain a “black box”. One problem is that the outcome is hard to measure in a complete way. Even though student achievement and some other measures are used in the present research, such measures do not capture the full set of the cognitive and non-cognitive skills of the students. Secondly, it is by no means clear that the objective of the decision-makers (e.g., school principals and local governments) is to produce skills in a least cost way. One interpretation of the seemingly weak link between resource use in schools and student achievement found in the literature (see below) is that the decision-makers take other objectives into account. In addition, it is in general not easier to estimate the effect of skills on costs than to estimate the effect of resource use on achievement. In the literature on the latter, the empirical results are mixed even with reasonable credible identification strategies.
Even though school level data hardly can be used to estimate cost functions, they may yield other interesting insight. Models of costs that do not include outcome measures can be seen as reduced form models in the sense that the same factors determine both the costs and the outcomes.
One issue that has received considerable attention in the literature is economies of scale. Andrews et al. (2002) review the literature from the US at the school district level and conclude that costs per student are minimized for about 2000–6000 students in the school district. For England, Taylor and Bradley (2000) use school-level data on staff hours per student. They find that the cost-minimizing school size is around 1600, which is about the size of the largest schools in England. They also find a strong negative effect on spending per student of school capacity utilization. Falch et al. (2008) finds that teaching hours per student in Norway is sharply decreasing for school size up to about 200 students in a model only utilization variation across schools within the same local government. They conclude that the optimal school size is above 800 students, which is largest school size in the country.
Both Taylor and Bradley (2000) and Falch et al. (2008) include characteristics of the student body composition in their models. Both papers find that the teacher intensity for students with special needs is about twice the average intensity, and that the teacher intensity is larger for minority students than for other students. The effect on costs of minority students is found to be larger in Norway than of having English as the second language in England.
These studies implicitly estimate parameters of the models used to allocate funding across schools. Little evidence exists, however, on whether funding models and school size have an impact on student achievement. Some education production function studies have simply included school size as a control variable, and the overview in Andrews et al. (2002) shows that the estimated effect of this variable is mixed. Berry and West (2008) use a more credible identification. They utilize that school consolidation in the US has occurred in different time periods in different states. Their findings indicate that increasing school size by school consolidation reduces both the years of completed education of the students and the return to education in terms of earnings.
The analysis of de Haan et al. (2011) is related both to school size, school choice, and competition. They exploit a change in the funding model of schools in the Netherlands to estimate the effect of the number of schools in a given geographical area (municipality) on student achievement. Dutch schools are funded by the state according to a formula based on the number of students enrolled. The same rule applies for private schools (mainly religious schools), which is in majority, and public schools. There is no additional funding from local governments, no tuition fees, and parents can freely choose the school for their child. The formula determining the minimum school size in order to be eligible for government funding changed in 1994. Both the new and the old formula were related to the density of the population, but the new formula implied that the minimum school size increased for almost all school locations. As a result, some schools closed, some school merged, and the average school size increased. At the same time, since the number of service providers decreased, the competition in the market decreased. de Haan et al. find that the reduction in the number of schools induced by the change in the funding formula significantly increased student achievement. They interpret the finding as evidence of scale economies; educational quality is increasing in school size. The economies of scale dominates potential negative effects of reduced school competition.
The apparently different effect of school size in Berry and West (2008) and de Haan (2011) must be interpreted with respect to the variation in school size they use for identification. Berry and West use variation around mean school size while de Haan et al. exploit variation at the bottom of the distribution of school size.
2.3. Centralized vs. decentralized finance
Should financing of education be the responsibility of central or local governments? The share of funds towards primary and secondary education provided by the central government varies greatly across the European countries. According to OECD (2010), central government funds financed over 70 percent of the spending in primary and secondary education in Austria, Ireland, Italy, Luxembourg, the Netherlands, the Slovak republic and Slovenia. On the other hand, central government funds finance less than 30 percent in Belgium, the Czech Republic, Germany, Iceland, Norway, Poland, Spain, Switzerland and UK. For the latter countries, the finance responsibility is mainly at the regional level in Belgium, Czech Republic, Germany and Spain, but at the local level in Iceland, Norway, Poland and UK. It must also be noted that even though the funds formally are raised at the local level, the local tax rates may be regulated by the central government.
Oates (1972) argues that more decentralized political decision-making allows better adjustment of supplies to locally heterogeneous demands. For services without spillovers across local governments, decentralized decisions contributes to allocation efficiency. There are several other arguments for decentralisation put forward (Oates, 2005). For example Seabright (1996) argues that the main advantage of decentralization is that it makes it is easier to hold decision-makers accountable. Hoxby (1999) provides a related argument. In her model, public school efficiency is higher with decentralized finance by local property taxes than with centralized finance since mobility decisions of the citizens reveal important information on local preferences and work as a discipline device for school providers. With property taxation, Tiebout mobility generates verifiable demand information that is useful in order to manage the productivity of local schools, information which is costly or impossible to gather with centralized finance. Property taxation links school quality to school finance. Fischel (2001) notes that with property taxation adults without children at school have an interest in schools using their resources efficiently since education quality is capitalized in house prices.
In the US, primary and secondary education is traditionally financed by local property taxes. During the latest decades, state grants have become increasingly important to a large part because of legislative decisions that find local finance unconstitutional. Hoxby (2001) argues that this school finance equalization reform is “probably the reform that has most affected American schools over the past 30 years” (p. 1189).
2.3.1. Education spending
A court case in California in 1971, which ordered the state to develop a system where school spending did not depend on school district wealth, launched the school finance reform. The following reform in California has motivated a large research literature. Fernández and Rogerson (1999) use a model where the quality of education is a
function solely of spending per student, and the families care about the quality of education in addition to consumption, to compare pre-reform and post-reform school finance systems. In the model, education finance and tax decisions are taken according to a majority rule (the median voter model). They conclude that the distribution of spending is more equal with centralized finance, mainly at the cost of reduced spending in wealthy school districts. Only a small fraction of the school districts will experience significantly higher spending under centralized finance than under local finance.
Silva and Sonstelie (1995) clarify the mechanisms. Given that citizens partly sort themselves across school districts with respect to their income (Tiebout mobility), the difference between median income and average income is larger at the state level than in the typical school district. In a majority rule model where the decisive voter has median income, the effect on overall spending of a change from decentralized to centralized finance depends on the size of an income effect relative to a price effect. The income effect is negative because the median income is further below the mean income, while the price effect is positive in a progressive tax system since the tax share of the median voter becomes lower. The relative size of these effects depends on the income distribution and the tax system. Little progressivity of the tax system and large income differences between school districts work in the direction of a large income effect and a small price effect, and thus low spending in a centralized system.
In a setting not relying on a particular decision making model, but merely on a description of the changes in school finance equalization schemes, Hoxby (2001) reaches a similar conclusion. She shows that whether equalization reforms decrease or increase spending depends on how grant and tax regimes are designed.
The finding for California that spending declined is, in fact, not replicated in studies comparing US states with decentralized and centralized finance. Murray et al. (1998) and Card and Payne (2002) find that court-ordered finance reforms reduced within-state inequality in spending by raising spending in the poorest school districts, mainly leaving spending in the richest districts unchanged. Thus, average spending per student is higher in states that increase the funding responsibility of the state. Card and Payne (2002) interpret this finding as a flypaper effect. They find that a one-dollar increase in state grants increases district education spending by 50-65 cents. Local taxes only slightly decline. The estimates are based on models that identify the grant effect by court rulings.
The US evidence clearly indicates that the geographical distribution of the education resources is lower under centralized finance than under decentralized finance. The evidence also shows that there is not a monotonic relationship between education finance systems and total resource use in education. Centralizing the funding
spending in a typical state. The design of local tax systems is probably important. As a response to the court lead reform in California, strict limitations were imposed on property taxes which essentially turned the local property tax into a state-wide tax (Sonstelie et al., 2000).
In Europe, education finance has traditionally been a responsibility for the central government. Using arguments for political accountability and efficiency, several countries, including Spain and Sweden, have initiated processes of decentralization. Sweden decentralized education finance in the early 1990s. The formal responsibility for primary and secondary education was transferred from the central government to the local governments. The local governments became financed by lump-sum grants and local income taxes (Björklund et al., 2004, Ahlin and Mörk, 2008). The wage setting of teachers was decentralized in the same period. Sweden is now considered to have one of the most decentralized schooling sectors in the OECD. Björklund et al. (2004) find that the distribution in education resources across local governments widened during the 1990s, in particular at the bottom of the distribution. This result is in line with the US evidence. Ahlin and Mörk (2008) investigate whether the local tax base and grants affected school resources differently as decentralization took place. They find that the reform did not change the effect of the local tax base on education resources per student. However, the effect of grants decreased. The results for the grants are probably at least partly related to the fact that the decentralization reform replaced matching grants with lump-sum grants.
2.3.2. Education performance
An important question is whether the finance system per se influences student performance. The incentive to local governments and schools to use resources efficiently may depend on the finance system as suggested by, e.g., Hoxby (1999) and Fischel (2001). Sonstelie et al. (2000) finds that the achievement of students in California relative to the rest of the US declined after the centralizing reform, but conclude that the decline cannot necessarily be blamed on state finance. Disentangle the effect of the school finance reform from the effect of other changes, including changes in overall resource use, is challenging.
Husted and Kenny (2000) use state level panel data on student achievement during the period 1987-1993, a period of finance reforms in many states. They find that state-induced spending equalization lowers student achievement. Hall (2007) uses a cross-section of school districts in Ohio to investigate a similar issue. He investigates whether the percent of school districts’ revenue from local sources and the share of the property that is classified as residential or agricultural have an impact on student achievement. The latter measure, conditional on the percent of local revenue, is strongly related to the
importance of property taxation. Hall (2007) finds that both measures are related to higher test scores.
Since a variety of unobserved factors are likely to be correlated both to the design of the equalization regimes, local tax policy, and student achievement, the result of studies such as Husted and Kenny (2000) and Hall (2007) must be interpreted with care. Card and Payne (2002) overturn this problem by investigating effects on the difference in achievement across groups of students with different level of parental education. They investigate whether school spending equalization within a state influence the achievement gap across the groups. They interpret their results tentatively, but the results indicate that centralization reforms reduce the achievement gap between students with highly-education and poorly-educated parents.
Hoxby (2001) look at high school dropout rates. Like Card and Payne (2002) she avoids investigating the effects of school finance reform directly. Hoxby estimates the effect of changes in the funding equalization schemes implicit in the state specific financial system, and finds weak effects. She concludes that the results “hint that equalization improves student achievement the most (perhaps only) in schools that would have very low spending if left to their own devices” (p. 1228). The results on average achievement by Husted and Kenny (2000) and on the distribution of achievement in Hoxby (2001) and Card and Payne (2002) are consistent only if equalization is achieved by reduced performance at the top of the student achievement distribution.
Fiva and Rønning (2008) exploit that in Norway only some local governments have property taxation. In their empirical period the law restricted property taxation to urban areas. Since levying property tax or not was a local government decision, they use official status as town in 1911 and settlement pattern in the local government as instruments. They find that existence of property taxation increases student achievement.
Falch and Fischer (2010) use repeated international comparative achievement tests (TIMSS and PISA) to investigate whether countries’ relative ranking is related to decentralization of the public sector. They measure decentralization as the percentage of sub-national government spending in general government spending, an indicator provided by the World Bank. The relationship is identified by comparing within-country changes in decentralization with changes in student achievement (the difference-in-differences approach). They find that countries decentralizing spending decisions tend to increase the performance of their schools. This result confirms the finding for Switzerland in Barankay and Lockwood (2007). In Switzerland, the cantons are responsible for education, but the share of education spending carried out by the counties in contrast to the cantonal level varies across the cantons. Barankay and
with legal local autonomy. In their empirical analysis using panel data, they find that the change in the share of the population obtaining university entry qualification is positively related to increased spending decentralization.
The Spanish constitution of 1978 laid down the foundations for decentralized responsibilities. According to Peña and Solé-Ollé (2009), the decentralization process, however, lasted for almost two decades. Historic autonomous regions, as for example Catalonia, got power over education policy right after the new constitution was established, while the decentralization of power to the other regions, as for example Madrid, was achieved much later. Peña and Solé-Ollé (2009) utilize the different timing of decentralization to identify the effect on educational attainment measured as the rate that stay-on in education after the end of compulsory education. They find no effect on the total stay-on rate. However, decentralization decreased the share of the cohorts enrolling in vocational training programs while the share of the cohort which stay-on to study Bachilleratoincreased.
2.4. Financial incentives for students
Financial rewards to students depending on achievement or attainment may be an efficient way to improve student performance. Several systems with conditional cash transfers have been in place in developing countries (see Oosterbeek and Patrinos, 2008, for an overview), but such incentive systems are less common in OECD countries. A conditional cash transfer program where introduced in 15 out of 150 local education authorities in England in 1999 for 16- to 18-year-olds who remained in full-time education (the Education Maintenance Allowance, EMA). The benefit was means-tested and only available to students with low-income parents. The maximum weekly payment was £40. Deardon et al. (2009) evaluate the pilot and find that the cash incentive substantially raised the stay-on rates after the end of compulsory education. In the first year (year 12 of education), the participation rate increased by about 4,5 percentage points, while the proportion receiving two years of education increased by about 6.7 percentage points. EMA became available nationwide in 2004.
Angrist and Lavy (2009) analyse a randomized trial towards low-achievers in Israeli high schools enrolled in the academic track. Passing the matriculation requirements in Israel qualify for post-secondary education. The maximum cash benefit for matriculation was about $1400. They find that the cash benefit generated a substantial increase in the matriculation rate for girls, but had no effect on boys. They argue that the achievement award clearly had positive social return.
Bettinger (2010) investigates a randomized experiment in a poor community in Ohio which provided cash payments to students in grades 3-6 for successful completion of their standardized testing. Each year the students take five different achievement tests.
Eligible students received $15 for each test on which they scored proficient or better. Bettinger find a strong effect of the cash incentive on test scores in mathematics, but no effect in reading, science, and social science.
2.5. Concluding remarks
Education spending is of interest at least because; (i) it constitutes a large part of government budgets; (ii) analyses of spending behaviour helps understand public sector decisions making; and (iii) spending must at some level be related to education quality. The literature on the latter issue, the effect of resource use on student achievement, is large and is not covered by the present paper. However, literature reviews, as offered by for example Hanushek (2003) and Krueger (2003), clearly indicate that there are great many circumstances in which increased spending do not improve student achievement. The literature reviewed here clearly indicates that centralized funding of education contributes to more equal education spending across local governments. On the other hand, decentralized finance by property taxation provides better incentives to improve school quality. There is potentially an efficiency-equity trade-off in the choice of finance scheme which is not easily avoided since the leakage of grants to tax reliefs seems to be highest when local governments have high tax autonomy.
The centralizing movement of education finance in the US equalized education spending partly because grants to a large degree where used on spending and not on reduced local tax rates. Such a “flypaper effect” seems to be depended on political structure and the efficiency of local taxes available for local governments, and in many states local property taxation became less efficient for the local governments as part of the centralizing reforms. It is a challenge to restore local incentives when the central government heavily redistributes resources. One important discipline device for school providers in a decentralized system is that inhabitants can choose to move to another local government. This indicates that some other discipline device is necessary for efficient provision of education under centralized finance. One such device that show some promise is school choice of some degree for students and parents. Another is financial incentives geared towards the students.
One may be concerned about the willingness and possibility to fund education in an aging society. The empirical evidence on the demand for education spending suggests that such a concern is misplaced. Even though the evidence clearly indicates that a larger share of elderly in society decreaseseducation spending per student, the evidence also shows that a smaller share of students in the population increases spending per student. Since a larger share of elderly and a smaller share of students in most cases go hand-in-hand, the effect on spending per student will tend to cancel out.
3. Financing higher education
Higher education is heavily subsidized in the European countries despite the fact that the private return to higher education is significant. Some arguments for subsidies can be found in the optimal tax literature. A redistributive income tax implies that the after-tax return to education is lower than the effect of education on productivity as reflected in gross wages. Income tax is a distortion to education decisions, and education subsidies can be used to restore the distortion. Intuitively, the more progressive the income tax, the higher is the optimal education subsidy to eliminate the tax-wedge on education (Bovenberg and Jacobs, 2005). Obviously, this argument alone cannot account for very low tuition fees, or even absence of tuition fees, as one observes in many European countries.
Education subsidies can also be used to correct for non-tax distortions of various kinds. One argument for subsidy is that education has positive externalities by fostering dissemination of knowledge as in endogenous growth models (Lucas, 1988, and Romer, 1990), but education may also have elements of signalling information yielding over-investment since the true return to education is lower than observed wage differences suggest (Spence,1973). Another issue is risk. The return to education is uncertain at the individual level, and thus risk-averse students will underinvest (Levhari and Weiss, 1974). The case that education hedges against labour market risks works in the opposite direction (Jacobs et al., 2009). Other popular arguments for education subsidies are; (i) imperfections in the capital market implies that families with low parental income cannot afford tuition fees and living costs of students; (ii) students are myopic in the sense that they give too low value to future outcomes relative to the outcomes close in time; (iii) students are debt averse; (vi) education contributes to income equality since the supply of high-skilled labour increases and the supply of low-skilled labour decreases.
The size of non-tax distortions influencing human capital investment decisions cannot be established by theoretical work. It is even uncertain whether externalities, risks, labour supply effects, and equity considerations work in the direction of high or low optimal subsidies. Stringent empirical work on these issues is, however, difficult. To a large part the relevant theoretical concepts are hard to observe, and universities and study programs are heterogeneous such that learning can only be indirectly observed in the labour market, but then with a time-lag.
The literature review here is restricted to empirical work on student behaviour. How do students react to increases in tuition fees and student aid? Evidence on whether student behaviour is rational in a sense that is in accordance with viewing education as investment in future income is important in order to judge the relevance of the arguments above. We end this section with a short consideration of those arguments.
3.1. Tuition fees
Underlying all theoretical work is that higher tuition fees, all else equal, reduce enrolment in higher education. The return to human capital investment is reduced since the investment cost increases. The US evidence supports the hypothesis. According to Heller’s (1997) review, a typical finding in the literature is that rising tuition by $100 decreases enrolment by 0.5-1.0 percentage points. Hilmer (1998) estimates an ordered probit model distinguishing between attendance at university, attendance at community college and non-attendance using a national representative sample of senior students at high school. The modelling framework considers average tuition by universities and colleges in the student’s home state as exogenous. The results indicate that an annual increase in tuition fees by $100 reduces university attendance by 1.0 percentage point and college attendance by 0.3 percentage points. Hilmer also find positive cross-price elasticities. Higher tuition at universities increases college attendance, while higher tuition in colleges increases university attendance.
Tuition fees were introduced in England in 1998, and has increased and made more flexible in the recent years. Apparently, the fee changes have been parts of major national reforms in ways that make it difficult to disentangle their effects from other reform elements (Greenaway and Haynes, 2003, Dearden et al. 2008). During this period, England has introduced income contingent students loans, changed the mechanisms for disbursing public funds across universities, introduced grants and bursaries to students, etc.
Two recent papers have analysed the introduction of tuition fees in some German Ländern in 2007. In Germany, universities are the responsibility of the Ländern, and private universities are almost non-existent. In 2005 the German constitutional court ruled that the federal ban on tuition fees was unconstitutional. Shortly after, 7 Ländern passed a law which introduced tuition fees at state universities. The fact that it is within-country variation in fees makes it easier to evaluate their effects. The fees were relatively low, typically 500 euro per semester. To avoid liquid constrained students, the same Ländern also introduced subsidized loan schemes.
Hübner (2009) uses data on all high-school graduates qualified for university studies in Germany and employs a difference-in-difference strategy. The strategy takes into account all factors of each Länder that do not change over time. He finds that the probability to enrol in higher education declined by 2.7 percentage points in the states introducing the fee. This is the relevant effect for the country, but is a lower bond for the Ländern since the high school students could choose to enrol in non-fee states. Taking mobility into account, Hübner finds an average treatment effect on the treated of 4.8 percentage points. The results are of the same magnitude as the US evidence.
German Ländern as control group since there historically has been very little student migration between east and west. They use application data for medicine and dentistry studies, and focus on whether mobility decisions are sensitive to the introduction of tuition. The findings indicate that introducing tuition reduces the probability of applying in the home Länder by about 2 percentage points.
A related issue is whether tuition fees have an impact on time to complete a degree. Gariabaldi et al. (2011) use data from Bocconi University, a private university in Milan. The Bocconi students in their data are assigned to one of 12 different tuition levels based on parental income. Gariabaldi et al. compare students close to the kinks in tuition levels in a discontinuity framework. They find that increasing fees by 1000 euro decreases the probability of late graduation with 5.2 percentage points.
3.2. Student aid
Student aid lowers the cost of higher education. Aid schemes are therefore expected to have exactly the opposite effect as tuition fees. Grant and scholarships are typically targeted towards students from low socioeconomic backgrounds. Dynarski (2000) investigates the introduction of a scholarship program in Georgia. In a difference-in-difference framework comparing Georgia to neighbouring states, she finds that an aid of $1000 increases college attendance rates by 4-6 percentage points. Dynarski (2002) summarizes the US literature and conclude that the best estimates are in the lower bound of the finding in Dynarski (2000). She also shows that subsidy does not have a consistently different impact on low-income individuals than others. The evidence is also mixed on whether the effect of aid is larger or smaller for students from disadvantage backgrounds than for students from the middle class.
Fredriksson (1997) performs a time-series analysis on the demand for higher education in Sweden. In his empirical period, 1967-1991, the interest rate subsidy on student loans and the amount of non-repayable grant varied. The reform in the study allowance scheme in 1989 contributes to this variation. He finds positive effects of predicted university wage premium, relative employment prospects, and the generosity of the study allowance scheme.
Nielsen et al. (2010) investigate the effect of a reform in the Danish student aid system in 1988. Prior to 1988, grants to students below 22 years of age were mean-tested. In addition, the 1998-reform increased the aid considerably. Nielsen et al. exploit that the mean-testing rule prior to the reform implied that the reform had different financial impact on seemingly identical individuals. They find that increasing the aid by about $1000 increases enrolment by 1.35 percentage points. They argue that one reason why they find a lower response than other studies might be that relatively large grants were in place already prior to the reform.
A change in the federal student aid scheme in Germany in 2001 both increased the number of students eligible for the aid and increased the amount of student aid for those eligible. Baumgartner and Steiner (2006) use a difference-in-difference strategy on data from the German Socio-Economic Panel Study to estimate the enrolment response to the reform. They find a positive but statistically insignificant effect of the reform. Steiner and Wrolich (2008) use the same data source, but relate the educational choice to the predicted student aid based on parents’ income. They find a significantly positive effect on enrolment into university of the same size as the US evidence.
Thus, the evidence indicates that students’ responses to tuition fees and aid are similar with opposite signs. This is in accordance with a hypothesis that students respond rational on economic incentives. However, some evidence point to some peculiarities in student behaviour. Student loan systems have relative low interest rates compared to all other kinds of loans. Given this feature, it seems to be surprisingly low up-take rates. In the Netherlands, only about 35 percent of available credit is taken out. Booij et al. (2011) conduct a field experiment where they manipulate the amount of information students have about the student loan conditions. They find that the information provided did increase the knowledge about the student loan conditions, but did not increase the take-up rate. This result indicates that lack of information about loan conditions is not a constraint. In a field study in the US, Bettinger et al. (2009) find that providing information about eligibility is not enough to induce students to apply for grants when the application is complex. The number of applications increases only when students receive direct help in the complicated application process.
3.3. Student loans
The private market for loan to studies is small and almost non-existing. Human capital cannot serve as collateral due to its illiquid and non-trading nature. If a borrower becomes unable to repay the loan, lenders have no security. Another imperfection in a private student loan market is the so-called adverse selection mechanism. Facing the possibility of large losses, lenders will try to discriminate between borrowers. They will offer a higher interest rate to students with the least expected future income and to students that cannot offer some security as for example the parents’ house. Thus, a traditional bank market will offer both inefficiently little student loans and inequity since students from low socioeconomic backgrounds will be faced with the worst loan conditions.
Consequently, there is a role for public student loan arrangements to overcome these imperfections. A public student loan system can offer a lower interest rate than private loans without providing subsidies. By covering the whole population of students, public loan arrangements can overcome both the problem of non-collaterals and the problem of adverse selection. This is not an argument for a low and subsidized interest rate that
yields overall losses of the system, but simply that public student loan arrangements will be more efficient and equitable than private alternatives.
Public student loan arrangements take different forms, see, e.g., Oosterbeek and Patrinos (2008). They can be of the mortgage-type, which in principle is equal to traditional loans and is used in the Nordic countries and the Netherlands, or they can be income contingent, such as in England. Mortgage-type loans are required to be repaid in the form of fixed instalment after graduation. Repayment of income contingent loans depends on achieved income. Students repay a percentage of their earnings after graduation. Strong arguments for the income contingent loans are provided by Barr (1993, 2004). One main feature of income contingent loans is that they to some extent insure against low future income. Loans are repaid when the return to education materialize. The risk associated with higher education is therefore reduced.
A third variant is graduate taxes. With graduate taxes, students receive a grant while studying, which they repay in the form of a special tax on income after graduation. Graduate taxes and income contingent loans have many similarities, but graduate taxes provide funding to all students whether they want it or not, and graduates with high income will end up paying more in tax than the grant they received. Graduate taxes are not used in any country (Oosterbeek and Patrinos, 2008).
Another issue is whether students are credit constraints in the sense that some do not enter higher education because they do not have the financial means to pay the tuition fees and the living costs. Research from the US indicates that family credit constraints in a child’s college-going years are of little importance in explaining a child’s enrolment in college. Carneiro and Heckman (2002) argue that only up to 8 percent of the total US population is credit constrained. The main reason for low enrolment rates of students from disadvantaged background is that they lack required qualifications for higher education studies. Family income during a child’s college-going years plays only a minor role in determining socioeconomic differences in college participation. With much lower tuition fees, it is likely even fewer individuals in Europe that do not take higher education because they cannot afford it in the short run. Indeed, this is the finding by Chowdry et al. (2010) for England.
3.4. Concluding remarks
Higher education is investment in future wage premiums. The empirical literature on student behaviour indicates that students mainly react in accordance with an investment model. Enrolment in higher education rises with student aid and declines with tuition. The responses to aid and tuition are of about the same magnitude, and the evidence indicates that student behaviour in this respect is independent of socioeconomic background.
